A Hard Look at Hardware and Software: Reducing the Cost of Modeling
Multi-core computer processing has made remarkable advances over the past three years and is now practically and affordably within reach of all modelers—not just those with big budgets. However, reaping the benefits of 32-, 64-, or 128-core CPU power in 2021 is not as straightforward as purchasing a dandy computer, installing software, and having at it. Gamers have clearly pushed the development of faster hardware but adapting that hardware for hydrologic and hydraulic modeling presents a new set of challenges. CPU temperatures, cooling methods, clock speeds, fan adjustments, memory protocols, and infinite combinations of process parameters require considerable attention and tuning. Yet even when the hardware is highly tuned, computing performance often ends up being throttled, for various reasons, to the point that less powerful computers are nearly as fast at delivering computational results. Tried-and-true H&H software tends to bottleneck itself when run on high-capacity CPUs and rarely achieves the full promise of the CPU.
Many online forum posts by engineers are looking to build faster computers to run programs such as HEC-RAS and SWMM. Whether running 2D simulations that take many hours to complete or 1D models that take several minutes—but must be run hundreds, or thousands, of times to complete a project—everyone appreciates the benefits of reduced run times.
GKY & Associates, Inc. investigated a limited number of hardware (and cloud-based) computing solutions and built a 32-core, 64-thread computer specifically to run 2D HEC-RAS and SWMM models as fast as possible. We benchmarked hardware and software performance and developed an appreciation for what it takes to reap significant gains in processing power. This presentation will highlight some of the challenges encountered and describe the somewhat surprising insights gained in the pursuit of faster computing.